Generative AI Operating Model Director — Consulting, Retail Banking

The mandate

The defined problem is that eleven GenAI pilots use different data, evaluation and approval practices, so the bank cannot compare value or move safely into production. The assignment does not include building the models.

The deliverable is a board-ready GenAI operating-model pack: decision rights, portfolio method, risk tiers, lifecycle controls, reusable delivery playbook, funding logic and a twelve-month adoption plan.

Milestone one is due 31 October 2026 with a pilot baseline and risk taxonomy; milestone two is due 15 January 2027 with the target model and two design-tested use cases; milestone three is due 15 March 2027 with approved playbooks, economics and transition backlog.

Acceptance requires every pilot mapped, control owners named, two use cases walked through end to end, and finance validation of benefits and run cost. The AI Steering Committee judges acceptance, with the Chief Risk Officer holding a control-readiness veto.

The bank provides model artefacts, vendor contracts, secure data access, pilot teams, finance assumptions and weekly sponsor access. Delayed access shifts milestone dates through documented change control rather than silent scope compression.

Why this is external work

Internal teams are invested in their own pilots and use incompatible success measures. The bank needs a neutral portfolio view plus specialist operating-model experience it does not retain full time. Independence also allows unresolved ownership disputes to be surfaced without favouring a business unit.

What you will own

• Establish the eleven-pilot baseline and reconcile competing value, risk and readiness definitions for milestone one.
• Classify use cases by customer harm, data sensitivity, autonomy and reversibility in the risk taxonomy.
• Design decision rights from idea intake through validation, production release and retirement for milestone two.
• Facilitate end-to-end design tests on one employee and one customer use case.
• Build portfolio economics covering inference, engineering, controls, human review and vendor commitments.
• Produce reusable evaluation, evidence and monitoring playbooks before milestone three.
• Present the transition backlog, accountable owners and twelve-month funding sequence for committee acceptance.

Candidate qualifications

• 18–22 years in bank technology, data, risk or enterprise-transformation leadership.
• Direct delivery of a governed AI operating model inside a regulated financial institution.
• Practical command of GenAI evaluation, retrieval, model risk, privacy and human-oversight patterns.
• Evidence of rationalising a fragmented innovation portfolio using comparable economics.
• Experience obtaining joint technology, finance and risk acceptance of enterprise controls.
• Ability to work within secure bank environments without exporting sensitive artefacts.

Non-negotiables

• Named senior director accountable for all three milestone outputs.
• No subcontracted model build, platform resale or vendor commission in the proposed fee.
• Bengaluru presence for fortnightly design sessions and every acceptance review.
• Agreement that missing client inputs trigger formal change control, not untested assumptions.

1. 49 words maximum. Describe the AI operating model you delivered in a regulated institution and the acceptance evidence used.
2. 49 words maximum. How would you baseline eleven pilots with incompatible value and risk measures in the first four weeks?
3. 49 words maximum. Which client inputs must be available by day ten to protect the 31 October milestone?